[MITgcm-support] Question about parallel run

Yuan Lian lian at ashimaresearch.com
Wed Jun 15 05:02:24 EDT 2011


No, -mods specifies the "code" directory that includes SIZE.h and/or 
other modified source codes.
For instance, if you want to run two simulations with different number 
of CPUs, then you can
make "code1" and "code2" directories, each contains its own SIZE.h with 
specified nPx and nPy.

You can't change the number of CPUs for a compiled mitgcmuv on the fly. 
You will need to compile
each model with its corresponding SIZE.h, that is, generate mitgcmuv for 
each SIZE.h.

Hope this clear things up.

Yuan


On 6/15/11 1:39 AM, Ali Nazemian wrote:
> So you meant i can specify different size.h file with -mods parameter 
> in genmake2 command? hereby i can add this to job scheduling file for 
> SGE and the problem will be solved.
> Regards.
>
> On Tue, Jun 14, 2011 at 11:10 PM, Martin Losch <Martin.Losch at awi.de 
> <mailto:Martin.Losch at awi.de>> wrote:
>
>     Ali,
>
>     you do not need different source code for each configuration, only
>     your SIZE.h needs to change when you build the executable
>     "mitgcmuv". I usually have different subdirectories when I want to
>     try different numbers of cpus. The makefile-procedures just links
>     the source code into the build directory so that you can use the
>     same source code for all of you runs, except for SIZE.h which
>     should then reside in a separate directory (specified by -mods at
>     the genmake2 step).
>
>     The code is based on Fortran77, so all memory is allocated
>     statically. I do not know of any plans to change that in the future.
>
>     Martin
>     On Jun 14, 2011, at 11:12 AM, Ali Nazemian wrote:
>
>     > Thanks a lot.
>     > So i should have a different source code per each run
>     configuration and use it for example with SGE as a job scheduler?
>     > Is it in developer road map to make MITgcm more portable? i mean
>     change parallel configuration to runtime instead of compile time?
>     > Regards.
>     >
>     > On Tue, Jun 14, 2011 at 6:10 PM, Stefano Querin
>     <squerin at ogs.trieste.it <mailto:squerin at ogs.trieste.it>> wrote:
>     > Hi Ali,
>     >
>     >
>     > Hi,
>     > I am wondering that should i recompile MITgcm each time that i
>     want to run that on our cluster with different number of processors?!
>     >
>     > Yes (as far as I know...), you have to recompile it.
>     >
>     >
>     > assume i want to run it on 10 processors, 16 processors , 20
>     processors and etc... should i change size.h file each time and
>     recompile my code?
>     >
>     > Yes.
>     > For more info:
>     >
>     http://mitgcm.org/public/r2_manual/latest/online_documents/node102.html
>     >
>     http://mitgcm.org/public/r2_manual/latest/online_documents/node181.html
>     >
>     >
>     > what if two or more user want to submit parallel job with MITgcm
>     with different number of processors?
>     >
>     > Just run them on separate folders...
>     > If many people want to perform several runs, you should also
>     have a job scheduler installed on your cluster to handle the job
>     queues.
>     > Ask your system administrator.
>     >
>     >
>     > have i need a different compiled code for each of them?!
>     >
>     > Yes, one executable for each kind of domain decomposition.
>     >
>     > Best Regards.
>     > --
>     > A.Nazemian
>     >
>     > Cheers,
>     >
>     > S.
>     >
>     >
>     > _______________________________________________
>     > MITgcm-support mailing list
>     > MITgcm-support at mitgcm.org <mailto:MITgcm-support at mitgcm.org>
>     > http://mitgcm.org/mailman/listinfo/mitgcm-support
>     >
>     >
>     > _______________________________________________
>     > MITgcm-support mailing list
>     > MITgcm-support at mitgcm.org <mailto:MITgcm-support at mitgcm.org>
>     > http://mitgcm.org/mailman/listinfo/mitgcm-support
>     >
>     >
>     >
>     > --
>     > A.Nazemian
>     > _______________________________________________
>     > MITgcm-support mailing list
>     > MITgcm-support at mitgcm.org <mailto:MITgcm-support at mitgcm.org>
>     > http://mitgcm.org/mailman/listinfo/mitgcm-support
>
>
>     _______________________________________________
>     MITgcm-support mailing list
>     MITgcm-support at mitgcm.org <mailto:MITgcm-support at mitgcm.org>
>     http://mitgcm.org/mailman/listinfo/mitgcm-support
>
>
>
>
> -- 
> A.Nazemian
>
>
> _______________________________________________
> MITgcm-support mailing list
> MITgcm-support at mitgcm.org
> http://mitgcm.org/mailman/listinfo/mitgcm-support

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mitgcm.org/pipermail/mitgcm-support/attachments/20110615/48b13eb4/attachment.htm>


More information about the MITgcm-support mailing list